Neutron, our AI engine, scored 96.75% on UC Berkeley's CyberGym benchmark. Learn more

Author

Sohaib Harraoui

Security Engineer

Sohaib Harraoui is a Security Engineer at Ostorlab, where he focuses on application security, attack surface discovery, and offensive vulnerability research. With a background in software engineering and systems, he investigates web and mobile ecosystems, analyzes how software assumptions break, and develops reproducible proof-of-concept exploits. His writing helps engineering and security teams understand real-world attack vectors, evaluate emerging threats, and build more resilient systems.

6 articles LinkedIn

False positives are an engineering-capacity problem, not only a scanner-quality problem. Learn how to measure their cost and evaluate the ROI of proof-of-exploit testing.

Security

The Fastest Way to Get an Audit-Ready Pentest Report for SOC 2 Compliance

Learn how B2B SaaS startups bypass the 4-week consultancy delay to generate audit-ready SOC 2 pen...

Sep 16, 2026

Security

Ostorlab Neutron on UC Berkeley CyberGym: 96.75% Verified Exploitation (1,458/1,507 Tasks)

A technical deep-dive into how Ostorlab Neutron reached a 96.75% verified exploit solved rate (1,...

Sep 15, 2026

Security

Autonomous Pentesting vs Traditional Pentesting

Compare traditional pentests, PTaaS and autonomous AI testing: where agents win on coverage and e...

Sep 08, 2026

More by Sohaib Harraoui

Breaking Down the Latest Version of GoPhish: Source-Code Assessment with Ostorlab Agentic Deep Scan

A technical assessment of the latest version of GoPhish that examines how the platform handles trust: identity, untrusted content, object ownership, credential lifecycle, and outbound requests. Source-code analysis with Ostorlab Agentic Deep Scan established the eight report-level findings, PoCs, and remediation priorities.

Jul 16, 2026

Exploit CVE-2026-42208: LiteLLM Unauthenticated SQL Injection via Bearer Token

A technical breakdown of CVE-2026-42208, a CVSS 9.3 critical unauthenticated SQL Injection vulnerability in the LiteLLM Proxy API. Improper parameterization of the Bearer token within raw SQL queries used for complex multi-table joins allows blind boolean-based timing attacks, enabling unauthenticated attackers to exfiltrate sensitive data including virtual API keys, user information, and LLM spend logs directly from the database.

May 22, 2026